ConductScience VR

Virtual radial arm maze

People move through a virtual eight-arm maze to collect rewards at the arm ends, as rats do in the radial arm maze, remembering which arms they have entered.

ConductScience Editorial · Published November 15, 2026 · 5 references

BLOCKED
Drawing · Virtual eight-arm maze
Task
Collect the rewards at the ends of four of the eight arms, without going back into an arm.1
Errors
Reentering an arm counts as a working memory error, and entering an arm that is never rewarded as a reference memory error.1
Strategies
A second version shows whether a person finds the arms by the landmarks or by a learned series of turns.2
Brain
Young adults who navigated by the landmarks had more gray matter in the hippocampus, and those who used turns had more in the caudate nucleus.3

What the virtual radial arm maze tests

The virtual radial arm maze is a computer version of the eight-arm maze that rodent labs use to test spatial memory.4 Spatial memory is often tested with different tasks in people and in other animals, and Astur and colleagues used the virtual maze to test people on a rodent task.4 In rodents, the maze relies on the prefrontal cortex and the hippocampus.1

The task scores two kinds of mistake.1 Spieker and colleagues define a reference memory error as entry to an arm that is never rewarded, and a working memory error as entry to an arm already entered.1 In their account, reference memory reflects hippocampal function and working memory reflects frontal function.1

A second design, the 4-on-8 virtual maze, shows which navigation strategy a person picks on their own.2 With a spatial strategy, a person memorizes how the landmarks around the maze relate to each other, which depends on the hippocampus.2 With a response strategy, a person memorizes a series of left and right turns from the start, which depends on the caudate nucleus.2

How a session runs

Studies run the virtual maze in two forms: the standard version, scored for errors, and the 4-on-8 version, built to reveal a strategy.1,2

The standard version

Four of the eight arms hold a reward, and the same four are rewarded on every trial.1 Every trial after the first starts in the middle of the maze.1 A pleasant tone sounds whenever a reward is found.1 Spieker and colleagues gave 10 trials, each with 5 minutes to find all four rewards.1 People steered with a joystick through a first-person view on a 17-inch monitor.1 Each trial’s time, distance, working and reference memory errors and rewards found were recorded.1

The 4-on-8 virtual maze

Stairs at the end of each arm lead down to a pit, so a person at the center cannot see which arms hold objects.2 Mountains, a sunset, two trees and a short wall surround the maze, and no landmark stands directly in front of a target arm.5,2 Every trial starts at the center, facing the same direction, and people move forward, left and right with a keypad.2 Bohbot and colleagues chose keys over a joystick so that people new to video games would not wander into arms by accident.3 Before the task, people practiced moving in a separate virtual room.5,3

  1. Collect the objects in four open arms

    Barriers block four arms, and the participant picks up the object at the end of each open arm, remembering which arms they visited.2 In trial type A, arms 1, 3, 4 and 6 are open.2

  2. Find the objects in the other four arms

    All eight arms then open, and the objects are in the four arms that were blocked, so the participant must avoid the arms already visited.2 Entering an arm with no object, or entering an arm twice, counts as an error, and the trial ends when all four objects are found.2 Performance is the number of errors in this second part, with all eight arms open.2

  3. Hide the landmarks in a probe trial

    The probe’s first part repeats trial type A, and then walls rise to hide the landscape, the trees are removed, and every arm holds an object.2 People who rely on the landmarks should now make more errors, while people who remember turns from the start should not.2 Bohbot and colleagues gave five trials in the order A, B, A, C, A, where C is the probe and B opens a different set of arms.2 Only people with no errors in the second part of an earlier trial went on to the probe.2

  4. Ask how the participant solved it

    Afterward, each participant describes how they solved the task.5,2 Counting the arms or labeling them with numbers or letters marks a response strategy, and using two or more landmarks marks a spatial one.2 Probe errors then confirm the report, since spatial learners are expected to make more of them.2

What the task has shown in people

Two strategies, and a shift with practice

Iaria and colleagues found that people split between the two strategies on their own.5 At the start, 23 of 50 people used the landmarks and 27 counted the arms.5 With practice, people shifted from the landmarks to counting, and by the end 36 of the 50 counted the arms.5 Those who used the landmarks made more errors on the probe trials, when the landmarks were hidden.5 A second experiment ran the same maze during functional MRI.5 Landmark users showed more activity in the right hippocampus early in training, and the others showed sustained activity in the caudate nucleus with practice.5

Brain structure

Bohbot and colleagues compared brain scans of 30 young adults who had run the maze.3 Spatial learners had more gray matter in the hippocampus and less in the caudate nucleus than response learners.3 The more gray matter a person had in one of these structures, the less they had in the other, which the authors read as competition between them.3

From childhood to old age

Bohbot and colleagues tested 599 healthy people aged 8 to 80 on the 4-on-8 maze.2 Use of a spatial strategy fell with age, from 84.4% of children to 46.3% of young adults and 39.3% of older adults.2 Young adults made fewer errors than older adults, and both did better than children.2 Strategy did not predict how well people learned the maze before the probe.2 The authors name repetition, reward and stress as factors that favor response strategies.2

Men and women

Astur and colleagues tested 61 undergraduates on the virtual radial maze, a virtual water maze and a mental rotation task.4 Men and women did not differ on the radial maze in working memory errors, reference memory errors or distance to the rewards.4 Men did better at mental rotation and at finding the hidden platform in the water maze.4 The authors concluded that the two mazes test spatial memory in different ways, and urged caution in using them interchangeably.4

Schizophrenia

Thirty-three people with schizophrenia and thirty-nine healthy controls each ran ten trials of the standard version.1 People with schizophrenia took more time, traveled farther, made more errors of both kinds and found fewer rewards.1 They did learn, but did not reach the level of the controls.1 Their reference memory errors fell across trials, while their working memory errors did not.1 Symptom severity was not related to performance in the maze.1

How it compares with the rodent radial arm maze

The virtual maze keeps the rodent layout, eight arms around a center with rewards at the ends of some of them.4,1 A rat or mouse runs a real maze, while a person moves through a scene on a screen with a joystick or keypad.1,2 Spieker and colleagues note that people solve the radial maze with spatial strategies like those of rodents.1

In the study by Iaria and colleagues, about half the people used each strategy, which Bohbot and colleagues note matches the proportions seen in rodents.5,3 With more training, 40% of the people who used the landmarks switched to counting, a response strategy, as rats had been shown to do.3 Rodent work has also found differences in the hippocampus and striatum between spatial and response learners, like the gray matter differences in people.3 Older rats also favor a response strategy more than young rats do, in a T-maze, which matches the shift seen in older adults.2

Lesions of the medial temporal lobe impair learning in the rodent radial maze, and functional MRI during the virtual maze shows hippocampal and frontal activity that agrees with the lesion work.1 Unlike a rat, a person can also say how they solved the maze, and the 4-on-8 studies sort people into strategy groups from that report, with probe errors as a check.5,2

Build a virtual radial arm maze with ConductScience VR

We write the task software in house and build each virtual radial maze to the protocol you plan to run.

A render of an eight-arm maze of black walkways on wooden bases, with an open well at the end of each arm, in an empty room with a wooden floor.
Earlier render · 2018An eight-arm maze with a well at the end of each arm, in a virtual room.
The same eight-arm maze standing on grass under an open sky.
Earlier render · 2018The same maze in an open field.
  • The number of arms, which arms hold rewards, the stairs or wells at the arm ends, and the landmarks around the maze
  • Reward searches, two-part trials with barriers on some arms, and probe trials that hide the landmarks
  • The start position and facing direction, time limits, and the tone at each reward
  • Each trial’s path, time, distance, rewards found, and working and reference memory errors, saved for analysis
  • A desktop screen with a keyboard or joystick, or a headset
  • The software alone, or a package with the headset and computer

We also build teaching versions, so students can test each other on the maze and count the errors.

Related pages

Questions

Is the virtual radial arm maze the same test as the rodent one?
It keeps the rodent design of eight arms and a center, with rewards at the ends of some arms.4,1 It is scored for two kinds of error: entering an arm that is never rewarded, a reference memory error, and returning to an arm already entered, a working memory error.1
What is the 4-on-8 virtual maze?
It is a version built to show whether a person navigates by the landmarks or by a learned series of turns.2 Four arms are open in the first part of each trial and all eight in the second, and a probe trial hides the landmarks.2
Does the virtual radial arm maze depend on the hippocampus?
It does for people who use the landmarks, who showed more right hippocampal activity in one study and more hippocampal gray matter in another.2,5,3 People who follow a learned series of turns rely more on the caudate nucleus.5,3
Do men and women score differently?
No sex difference appeared in the virtual radial maze, in either kind of error or in distance to the rewards.4 The same undergraduates did differ in the virtual water maze, where men found the hidden platform better.4
Can the virtual radial arm maze run in a headset?
Yes, we build it for a headset or for a desktop screen with a keyboard or joystick.

References

  1. Spieker EA, Astur RS, West JT, et al. Spatial memory deficits in a virtual reality eight-arm radial maze in schizophrenia. Schizophrenia Research. 2012. doi:10.1016/j.schres.2011.11.014
  2. Bohbot VD, McKenzie S, Konishi K, et al. Virtual navigation strategies from childhood to senescence: evidence for changes across the life span. Frontiers in Aging Neuroscience. 2012. doi:10.3389/fnagi.2012.00028
  3. Bohbot VD, Lerch J, Thorndycraft B, et al. Gray matter differences correlate with spontaneous strategies in a human virtual navigation task. The Journal of Neuroscience. 2007. doi:10.1523/JNEUROSCI.1763-07.2007
  4. Astur RS, Tropp J, Sava S, et al. Sex differences and correlations in a virtual Morris water task, a virtual radial arm maze, and mental rotation. Behavioural Brain Research. 2004. doi:10.1016/j.bbr.2003.08.024
  5. Iaria G, Petrides M, Dagher A, et al. Cognitive strategies dependent on the hippocampus and caudate nucleus in human navigation: variability and change with practice. The Journal of Neuroscience. 2003. doi:10.1523/JNEUROSCI.23-13-05945.2003

More virtual reality tests